I personally am getting better results with codex recently. Claude ($20 plan) honestly comes across as a total ai slop turd of an app (unreliable, frequent incidents, burns through the token after 2-3 prompts that just clinfinite loop doing nothing). Codex will iterate much faster.

It’s all anecdata, but I heavily used Claude over the last two weeks and found it very reliable. My company pays for tokens and I haven’t noticed incidents.

And yet so many people are having success with Claude Code. Perhaps you can learn from their experiences.